Synthetic Electroretinogram Signal Generation Using Conditional Generative Adversarial Network for Enhancing Classification of Autism Spectrum Disorder

TL;DR

This paper tackles the data scarcity and heterogeneity challenge in using electroretinogram (ERG) signals to classify autism spectrum disorder (ASD). It introduces a conditional GAN (CGAN) to generate synthetic ERG waveforms for ASD and control classes, and combines these with real data to train two classifiers: a Visual Transformer (ViT) on continuous wavelet transform (CWT) scalograms and a Time Series Transformer (TST) on raw time-domain ERG signals. The CGAN objective uses a conditional min–max loss $V(D,G)$ to produce class-consistent synthetic signals, and the CWT with three mother wavelets provides rich time–frequency representations for ViT, while TST leverages temporal attention. Results show that synthetic data augmentation improves classification performance, with a best $BA$ of $0.805$ for the TST at a particular flash strength and notable gains for ViT, validating the approach and suggesting wide applicability to other psychiatric conditions where ERG signals may be informative. The work enables dataset expansion, supports training of heavy models like transformers, and makes synthetic signals suitable for open distribution, contributing a novel methodology for ERG-based ASD detection and beyond.

Abstract

The electroretinogram (ERG) is a clinical test that records the retina's electrical response to light. The ERG is a promising way to study different neurodevelopmental and neurodegenerative disorders, including autism spectrum disorder (ASD) - a neurodevelopmental condition that impacts language, communication, and reciprocal social interactions. However, in heterogeneous populations, such as ASD, where the ability to collect large datasets is limited, the application of artificial intelligence (AI) is complicated. Synthetic ERG signals generated from real ERG recordings carry similar information as natural ERGs and, therefore, could be used as an extension for natural data to increase datasets so that AI applications can be fully utilized. As proof of principle, this study presents a Generative Adversarial Network capable of generating synthetic ERG signals of children with ASD and typically developing control individuals. We applied a Time Series Transformer and Visual Transformer with Continuous Wavelet Transform to enhance classification results on the extended synthetic signals dataset. This approach may support classification models in related psychiatric conditions where the ERG may help classify disorders.

Figures (5)

• Figure 1: Proposed method overview. The dataset was split into test and training subsets so that the generative model was not trained on data correlated to the test for a fair evaluation. Synthetic signals generated by the generative model were added to real data. Classification models were then trained on merged and unmerged datasets in the time-frequency (ViT) and time (TST) domains.
• Figure 2: Conditional generative adversarial network (CGAN) structure.
• Figure 3: Examples of the real and synthetically generated by CGAN ERG signals for (a) 0.114, (b) - 0.119, (c) - 0.367, (d) 1.204 flash strengths $(log\: cd.s.m^{-2})$. Solid color lines represent synthetic ERG signals, dashed lines are real signals. Blue color corresponds to control, red color to ASD signals.
• Figure 4: Overview of the Visual Transformer (ViT) architecture.
• Figure 5: Series of real and synthetic control ERG signals generated by CGAN and filtered using Butterworth low-pass filter.